Predictors of post-retirement mortality in athletes required to maintain very high body mass: a retrospective cohort study using AI-driven three-dimensional body reconstruction, 1865–1990

Background Elite athletes in weight-uncapped disciplines sustain very high body mass throughout their careers (body mass index (BMI) 35–50 kg/m 2 ) and show a two- to three-fold mortality excess, yet no post-retirement care framework exists. Prior sumo mortality analyses used career-entry as follow-up origin, introducing immortal time bias. We generated virtual three-dimensional body reconstructions from historical photographs using artificial intelligence (AI)-driven landmark extraction to examine whether a reconstructed waist circumference adds prognostic information on post-retirement mortality beyond BMI. Methods All wrestlers reaching the makuuchi division (1865–1990 birth cohort; source N = 978) were assembled from published compendia. Transverse abdominal diameter was measured on career-peak photographs calibrated against recorded height; sagittal diameter was recovered by a volume-conserving solver; virtual waist circumference (C3) was the perimeter of the resulting ellipse. Retirement-anchored Cox models adjusted for birth year, retirement age and rank entered one body-size measure at a time; cause-specific models used follow-up from 1990 with multiple imputation of undocumented causes. Standardised mortality ratios (SMRs) used age × calendar-year-specific Japanese male rates. Primary cohort: birth < 1970 ( n = 604; 587 with a calibrated photograph). Results The primary cohort had 467 deaths over median 31.1 years follow-up; median BMI 36.2 kg/m 2 , median C3 136 cm. C3 rose with rank and correlated with BMI ( r = 0.95). C3 per standard deviation (SD) (hazard ratio (HR) 1.47, 95% confidence interval (CI) [1.29–1.69]; Harrell’s C 0.675) and BMI per SD (HR 1.38, 1.22–1.55; C 0.672) showed similar prognostic performance and could not be separated. The photographic diameter alone did not predict mortality (HR 1.08, 0.95–1.22) and added nothing to BMI (HR 1.03). Elite rank (HR 1.08) and career length (HR 0.98) showed no association. From 1990, with imputation, BMI predicted cardiometabolic death (HR 1.42, 1.08–1.87) but not neoplastic death (HR 1.07, 0.73–1.56), and C3 showed the same pattern (cardiometabolic HR 1.50, 1.08–2.07; neoplastic HR 1.20, 0.77–1.86); elite rank predicted neither. SMRs were 1.89 (1.72–2.07) overall, 1.84 for pre-1950 births and 2.57 for post-1950 births. Discussion Sumo wrestlers experienced about twofold all-cause mortality versus matched Japanese males across all historical strata. A waist circumference reconstructed from photographs, height and weight predicted mortality—specifically cardiometabolic mortality—as well as BMI, but its prognostic information derived from recorded body mass: the photographic component added none. Because obesity onset was unrecorded, the reconstruction is unvalidated, and no post-retirement data exist, the null findings for career length and photographic shape are not evidence that obesity duration or abdominal fat distribution is unrelated to mortality.